Test levels and risk-based testing · Question 60 of 60
Which mappings of risk and area along the ML workflow are correct?
4 answers are correct. Hard
Show answer
Answer
- B Biased training data: input data
- C Susceptibility to adversarial examples: model
- E Data pipeline defects: input data
- F Poor selection of evaluation approach: development
Explanation
Biased or unrepresentative training data and data pipeline defects are product risks in the input data area. Sub-optimal algorithm or evaluation selection and framework security vulnerabilities are project risks in the development area. Overfitting, missed performance measures and susceptibility to adversarial examples are product risks in the model area.